File size: 1,387 Bytes
36e528d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 | # TaskTrove v3.42 five-model data-quality tracker
All five model campaigns are sealed against the 103-source canonical TaskTrove v3.42 inventory at commit `3e96fe6464ce5ab6209e98801caab29b4a1fe87a`. Deprecated sources and datasets with fewer than 300 tasks are excluded.
| Model | Harness | Scored sources | Campaign tracker |
|---|---|---:|---|
| Qwen3-Coder-30B-A3B-Instruct | Terminus-2 | 103/103 | [`campaigns/qwen3-coder/TRACKER.md`](campaigns/qwen3-coder/TRACKER.md) |
| Qwen3.5-122B 131k | OpenCode | 103/103 | [`campaigns/qwen35-122b/TRACKER.md`](campaigns/qwen35-122b/TRACKER.md) |
| GLM-5.2 | Terminus-2 | 103/103 | [`campaigns/glm52/TRACKER.md`](campaigns/glm52/TRACKER.md) |
| Qwen3.6-35B-A3B | Terminus-2 | 103/103 | [`campaigns/qwen36-gemma4/TRACKER.md`](campaigns/qwen36-gemma4/TRACKER.md) |
| Gemma-4-26B-A4B-it | Terminus-2 | 103/103 | [`campaigns/qwen36-gemma4/TRACKER.md`](campaigns/qwen36-gemma4/TRACKER.md) |
The Qwen3.6 self-instruct cell and Gemma-4 OpenSWE cell are sealed by user override below the normal infrastructure gate. Their campaign tracker preserves the actual scoreable denominators and means.
## Reports
- [Five-model TaskTrove v3.42 comparison](reports/unified_reward_deltas_v342.html)
- [Five-model reward table](reports/unified_reward_table_v342.csv)
- [Pi versus Terminus-2 matched comparison](reports/pi_vs_terminus2_tasktrove_matched.csv)
|